Environmental Sounds Recognition
O. Aranda-Uribe, Mariko Nakano-Miyatake, Hector Manuel Perez-Meana · Telecommunications and Radio Engineering · 2006
This paper describes an environmental sounds recognition system using LPC-Cepstral coefficients as feature vectors and an artificial neural network backpropagation as recognition method. LPC-Cepstral data are totally dependents of the sound-source from which are computed. This system is evaluated using a database containing files from four different sound-sources under a variety of recording conditions. The training patterns used in the network-training ad testing processes, are extracted from the Discrete Fourier transform magnitude of the LPC-Cepstral matrices. The global percentages of classification obtained in the network-testing process are 98.2% and 96.8%. Basically the idea here is to apply the techniques found in speech recognition systems to an environmental sounds recognition system.